A federated learning method and apparatus based on feature calibration

Abstract

The application discloses a kind of federal learning method and device based on feature calibration, by obtaining demander submodel and provider submodel;First feature data provided by business demander and second feature data provided by data provider are obtained, and calibration model is generated according to the training of first feature data and second feature data;According to the feature calibration relationship of calibration model, demander submodel and provider submodel, joint training is carried out to obtain joint model;In response to the data query request of business demander, the first entry model feature data of business demander is selected, and the second entry model feature of data provider is determined according to the feature calibration relationship of calibration model, and the query result is obtained according to the first entry model feature data and second entry model feature Call joint model.The application applies feature calibration strategy in the process of training joint model, avoids the intervention of data provider data in prediction stage, and solves the problem of data security hidden danger.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a federated learning method and device based on feature calibration. Background Art

[0002] When optimizing and improving the performance of their own industry models, various industries often need to integrate user data from multiple parties. Currently, there are two main ways for all parties to share user personal information data: direct connection mode and federated learning (FL) mode. Among them, the direct connection mode is that the data provider directly provides the credit data interface to the demander, and the demander can directly access, obtain and use the personal data shared by the data provider, such as user identity information, basic information, behavioral information, and personal profile scores. The federated learning mode is to complete multi-party joint machine learning training through the circulation and processing of intermediate encrypted data without the local raw data being stored. Through federated learning, multiple users can conduct cross-industry business cooperation (such as joint risk control, intelligent marketing, etc.) without storing or sharing data. In addition, during the modeling process, no sensitive data is circulated, and only intermediate results such as encrypted calculation loss functions and gradients are exchanged, realizing multi-party joint secure modeling. Figure 1 A schematic diagram of a conventional federated learning model in the prior art is shown. Figure 1 As shown in the figure, the federated learning model is a distributed model, which includes two stages: model training and business application. In both the model training stage and the business application stage (i.e., application model prediction), the data provider needs to provide its own data for model training and prediction; in the business application stage, the data provider needs to provide probability scores transformed according to individual features based on its own sub-model to calculate the total score.

[0003] However, in the direct connection model, the data provider's data must be copied and transferred to one party or aggregated with a third party for data fusion before proceeding to the next step of data modeling. This model creates chaos as business demanders (such as banks or financial institutions) independently access personal information data, which poses data security risks. For data providers, data released from the database is easily illegally obtained by others. The disadvantage of the federated learning model is that the data provider's data still needs to be involved in the prediction phase, which also poses data security risks. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed to provide a method and apparatus for federated learning based on feature calibration that overcomes the above problems or at least partially solves the above problems.

[0005] According to one aspect of the present invention, a federated learning method based on feature calibration is provided, comprising:

[0006] Obtain a demander sub-model and a provider sub-model; wherein the demander sub-model is trained by the business demander based on the business demander's data, and the provider sub-model is trained by the data provider based on the data provider's data;

[0007] Acquire first feature data provided by the business demander and second feature data provided by the data provider, and train and generate a calibration model based on the first feature data and the second feature data;

[0008] Performing joint training on the characteristic calibration relationship of the calibration model, the demander sub-model, and the provider sub-model to obtain a joint model;

[0009] In response to the data query request of the business demander, the first input feature data of the business demander is selected, and the second input feature of the data provider is determined based on the feature calibration relationship of the calibration model, and the joint model is called based on the first input feature data and the second input feature to obtain the query result.

[0010] According to another aspect of the present invention, a federated learning apparatus based on feature calibration is provided, comprising:

[0011] A sub-model acquisition module is used to acquire a demander sub-model and a provider sub-model; wherein the demander sub-model is obtained by training the business demander based on the business demander's data, and the provider sub-model is obtained by training the data provider based on the data provider's data;

[0012] A calibration model training module is used to obtain the first feature data provided by the business demander and the second feature data provided by the data provider, and to train and generate a calibration model based on the first feature data and the second feature data;

[0013] A joint model training module, configured to perform joint training based on the characteristic calibration relationship of the calibration model, the demander sub-model, and the provider sub-model to obtain a joint model;

[0014] An application module is used to respond to the data query request of the business demander, select the first input feature data of the business demander, and determine the second input feature of the data provider based on the feature calibration relationship of the calibration model, and call the joint model based on the first input feature data and the second input feature to obtain the query result.

[0015] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0016] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned federated learning method based on feature calibration.

[0017] According to another aspect of the present invention, a computer storage medium is provided, wherein the storage medium stores at least one executable instruction, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned federated learning method based on feature calibration.

[0018] According to a method and apparatus for federated learning based on feature calibration of the present invention, a demander sub-model and a provider sub-model are obtained; wherein the demander sub-model is obtained by training the business demander based on the business demander's data, and the provider sub-model is obtained by training the data provider based on the data provider's data; first feature data provided by the business demander and second feature data provided by the data provider are obtained, and a calibration model is generated based on the first feature data and the second feature data; a joint training is performed on the demander sub-model and the provider sub-model based on the feature calibration relationship of the calibration model to obtain a joint model; in response to a data query request from the business demander, the first input feature data of the business demander is selected, and the second input feature of the data provider is determined based on the feature calibration relationship of the calibration model, and the joint model is called based on the first input feature data and the second input feature to obtain a query result. The present invention applies a feature calibration strategy to the demander sub-model and the provider sub-model during the training of the joint model, avoids the intervention of the data provider's data in the prediction stage through feature calibration, and applies the feature calibration strategy during the training of the joint model to avoid the intervention of the data provider's data in the prediction stage, thereby solving the data security risk problem, effectively avoiding unnecessary data flow, and reducing the data cost of the business demander to a certain extent.

[0019] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0021] Figure 1 A schematic diagram of a conventional federated learning model in the prior art is shown;

[0022] Figure 2A flow chart of a federated learning method based on feature calibration provided by an embodiment of the present invention is shown;

[0023] Figure 3 A schematic diagram illustrating a model training process and an application process of a federated learning method based on feature calibration provided by an embodiment of the present invention is shown;

[0024] Figure 4 A schematic diagram of a calibration model training process according to an embodiment of the present invention is shown;

[0025] Figure 5 A schematic diagram of a calibration model generation process according to an embodiment of the present invention is shown;

[0026] Figure 6 A schematic diagram of a first calibration model generation process provided by an embodiment of the present invention is shown;

[0027] Figure 7 A schematic diagram of a second calibration model generation process provided by an embodiment of the present invention is shown;

[0028] Figure 8 A schematic diagram of the structure of a federated learning device based on feature calibration provided by an embodiment of the present invention is shown;

[0029] Figure 9 A schematic structural diagram of a computing device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0030] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0031] Figure 2 A flow chart of an embodiment of a federated learning method based on feature calibration of the present invention is shown as follows: Figure 2 As shown, the method includes the following steps:

[0032] Step S210: Obtain the demand side sub-model and the supply side sub-model.

[0033] Figure 3 A schematic diagram of a model training process and application process of a federated learning method based on feature calibration provided by an embodiment of the present invention is shown, Figure 3The demander sub-model shown is trained by the service demander using a federated learning tool based on the service demander's data. The provider sub-model is trained by the data provider using a federated learning tool based on the data provider's data. In this embodiment, the service demander data is used as sample data to train the demander sub-model through the federated learning tool, and the service provider data is used as sample data to train the provider sub-model through the federated learning tool. The model is then tuned using the model tuning gradient value, and distributed training of the joint model is performed.

[0034] Step S220: Acquire first feature data provided by the business demander and second feature data provided by the data provider, and train and generate a calibration model based on the first feature data and the second feature data.

[0035] Specifically, in this embodiment, the demander sub-model and the provider sub-model are jointly modeled through machine learning, and the training processes of the demander sub-model and the provider sub-model are independent of each other. That is to say, the business demander data and the data provider data can be modeled locally respectively; therefore, only the intermediate parameters of the demander sub-model and the provider sub-model need to be exchanged in the joint model, such as the loss function and the gradient value, etc., to avoid data circulation throughout the process, reduce the risk of data interaction, and improve data security.

[0036] In an optional manner, step S220 further includes: taking the first feature data and the second feature data as sample data, and dividing the sample data into training set data and validation set data; training based on the training set data to obtain a calibration model; and validating the calibration model based on the validation set data.

[0037] Figure 4 FIG. 4 shows a schematic diagram of the calibration model training process provided by an embodiment of the present invention. Figure 4 As shown, the first feature data and the second feature data are used as sample data in the training stage, and the sample data are divided into calibration model training set data and calibration model validation set data in proportion. A calibration model is generated according to the calibration model training set data, and the calibration model is validated based on the validation set data to ensure the prediction accuracy of the calibration model; for example, 70% of the sample data can be used as training set data and 30% as validation set data.

[0038] In an optional manner, the calibration model includes: a first calibration model, a second calibration model and a third calibration model; step S220 further includes: calculating the prediction accuracy of the first calibration model, the second calibration model and the third calibration model; sorting the first calibration model, the second calibration model and the third calibration model from high to low according to the prediction accuracy, and determining the final calibration model based on the sorting result.

[0039] Figure 5FIG. 4 shows a schematic diagram of a calibration model generation process according to an embodiment of the present invention. Figure 5 As shown, the first feature data provided by the business demander is used as the data required for prediction, and the second feature data provided by the data provider is used as the data to be predicted. Three calibration models are generated by training through different algorithm models, and the model with the best effect among the three calibration models is taken as the final calibration model and put into the application stage. The second input feature of the data provider can be obtained by inputting the first input feature data of the business demander.

[0040] Among them, multiple calibration models include: a first calibration model; step S220 further includes: performing clustering operations on the sample data in the training set data to obtain a first cluster group of the business demander and a second cluster group of the data provider; extracting the common feature data in the first feature data and the second feature data in the sample data, constructing a cluster index based on the common feature data, and training and generating a first calibration model based on the cluster index, and the cluster index is used to query the second cluster group of the data provider based on the first cluster group of the business demander.

[0041] Figure 6 FIG. 4 shows a schematic diagram of a first calibration model generation process according to an embodiment of the present invention. Figure 6 As shown, based on the shared feature hierarchical clustering algorithm, the sample data in the training set data is clustered based on the features of both the data provider and the business demander, obtaining the first cluster of the business demander and the second cluster of the data provider. The shared feature data in the first feature data and the second feature data in the sample data are extracted, and a cluster index is constructed based on the shared feature data. The first calibration model is trained and generated based on the cluster index. The cluster index is used to query the second cluster of the data provider based on the first cluster of the business demander. For example, based on the age of the data provider and the business demander, data provider cluster 1 and business demander cluster 1 are generated respectively; based on the gender of the data provider and the business demander, data provider cluster 2 and business demander cluster 2 are generated respectively; based on the educational background of the data provider and the business demander, data provider cluster 3 and business demander cluster 3 are generated respectively; in addition, shared features can also include information such as number segments and address areas.

[0042] Among them, the calibration model includes: a second calibration model; step S220 further includes: using the second feature data in the training set data as the prediction target, using a machine learning algorithm, and performing prediction target fitting training based on the first feature data in the training set data to obtain the second calibration model.

[0043] Specifically, machine learning algorithms can include decision tree algorithms, Figure 7 FIG. 4 shows a schematic diagram of a second calibration model generation process according to an embodiment of the present invention. Figure 7As shown, the second feature data provided by the data provider in the training set is used as the prediction target, and the prediction target fitting training is performed according to the first feature data provided by the business demand party in the training set through a decision tree algorithm, etc., to obtain multiple models. For example, the important features in the second feature data are used as prediction targets, such as age, call time, gender and other important features, respectively as prediction targets, and prediction target fitting training is performed according to the first feature data provided by the business demand party in the training set, to obtain multiple models, and these multiple models can be constructed into a second calibration model.

[0044] Among them, the calibration model includes: a third calibration model; step S220 further includes: extracting relevant feature data between the first feature data and the second feature data in the training set data; taking the second feature data as the missing value, completing the second feature data based on the relevant feature data and the preset algorithm, and using the first feature data to perform feature calibration, and training to generate the third calibration model.

[0045] Among them, the preset algorithms include: mean interpolation algorithm, maximum likelihood estimation algorithm and multiple interpolation algorithm.

[0046] Step S230: performing joint training based on the feature calibration relationship of the calibration model, the demander sub-model, and the provider sub-model to obtain a joint model.

[0047] Specifically, a calibration relationship between the first feature data of the business demander and the second feature data of the data provider is generated based on the calibration model, and the demander sub-model and the provider sub-model are combined for joint training to obtain a joint model.

[0048] Step S240: In response to the data query request from the business demander, the first input feature data of the business demander is selected, and the second input feature of the data provider is determined based on the feature calibration relationship of the calibration model, and the joint model is called based on the first input feature data and the second input feature to obtain the query result.

[0049] Specifically, in the application stage, in response to the data query request of the business demander, the business demander selects the first input feature data of the business demander based on the training results, and backfills the second input feature of the data provider after prediction based on the calibration model. After the feature fusion, joint predictive modeling can be carried out, and the joint model is called based on the first input feature data and the second input feature to obtain the query result.

[0050] The method of this embodiment obtains a demander sub-model and a provider sub-model; wherein the demander sub-model is obtained by training the business demander based on the business demander's data, and the provider sub-model is obtained by training the data provider based on the data provider's data; obtains first feature data provided by the business demander and second feature data provided by the data provider, and generates a calibration model based on the first feature data and the second feature data; jointly trains the demander sub-model and the provider sub-model based on the feature calibration relationship of the calibration model to obtain a joint model; responds to a data query request from the business demander, selects the first input feature data of the business demander, and determines the second input feature of the data provider based on the feature calibration relationship of the calibration model, and calls the joint model based on the first input feature data and the second input feature to obtain a query result. This method applies a feature calibration strategy to the demander sub-model and the provider sub-model during the training of the joint model, avoids the intervention of the data provider data in the prediction stage through feature calibration, and applies the feature calibration strategy during the training of the joint model to avoid the intervention of the data provider data in the prediction stage, thereby solving the data security risk problem, effectively avoiding unnecessary data flow, and reducing the data cost of the business demander to a certain extent.

[0051] Figure 8 FIG1 shows a structural diagram of an embodiment of a federated learning device based on feature calibration according to the present invention. Figure 8 As shown, the device includes: a sub-model acquisition module 810, a calibration model training module 820, a joint model training module 830 and an application module 840.

[0052] The sub-model acquisition module 810 is used to obtain the demander sub-model and the provider sub-model; wherein the demander sub-model is obtained by training the business demander based on the business demander data, and the provider sub-model is obtained by training the data provider based on the data provider data.

[0053] The calibration model training module 820 is used to obtain the first feature data provided by the business demander and the second feature data provided by the data provider, and to train and generate a calibration model based on the first feature data and the second feature data.

[0054] In an optional manner, the calibration model training module 820 is further used to: use the first feature data and the second feature data as sample data, and divide the sample data into training set data and verification set data; train based on the training set data to obtain a calibration model; and verify the calibration model based on the verification set data.

[0055] In an optional manner, the calibration model includes: a first calibration model; the calibration model training module 820 is further used to: perform clustering operations on the sample data in the training set data to obtain a first cluster group of the business demander and a second cluster group of the data provider; extract the common feature data from the first feature data and the second feature data in the sample data, construct a cluster index based on the common feature data, and generate a first calibration model based on the cluster index training, and the cluster index is used to query the second cluster group of the data provider based on the first cluster group of the business demander.

[0056] In an optional manner, the calibration model includes: a second calibration model; the calibration model training module 820 is further used to: use the second feature data in the training set data as the prediction target, use a machine learning algorithm, perform prediction target fitting training based on the first feature data in the training set data, and obtain a second calibration model.

[0057] In an optional manner, the calibration model includes: a third calibration model; the calibration model training module 820 is further used to: extract relevant feature data between the first feature data and the second feature data in the training set data; use the second feature data as the missing value, complete the second feature data based on the relevant feature data and the preset algorithm, and use the first feature data to perform feature calibration to train and generate the third calibration model.

[0058] In an optional manner, the preset algorithms include: a mean interpolation algorithm, a maximum likelihood estimation algorithm, and a multiple interpolation algorithm.

[0059] In an optional manner, the calibration model training module 820 is further used to: calculate the prediction accuracy of the first calibration model, the second calibration model and the third calibration model; sort the first calibration model, the second calibration model and the third calibration model from high to low according to the prediction accuracy, and determine the final calibration model based on the sorting results.

[0060] The joint model training module 830 is used to perform joint training based on the feature calibration relationship of the calibration model, the demander sub-model and the provider sub-model to obtain a joint model.

[0061] Application module 840 is used to respond to the data query request of the business demander, select the first input feature data of the business demander, and determine the second input feature of the data provider based on the feature calibration relationship of the calibration model, and call the joint model based on the first input feature data and the second input feature to obtain the query result.

[0062] The device of this embodiment obtains a demander sub-model and a provider sub-model; wherein the demander sub-model is trained based on the demander's data, and the provider sub-model is trained based on the data provider's data; obtains first feature data provided by the demander and second feature data provided by the data provider, and generates a calibration model based on the first feature data and the second feature data; jointly trains the demander sub-model and the provider sub-model based on the feature calibration relationship of the calibration model to obtain a joint model; responds to a data query request from the demander, selects the first input feature data of the demander, determines the second input feature of the data provider based on the feature calibration relationship of the calibration model, and calls the joint model based on the first input feature data and the second input feature to obtain a query result. The device applies a feature calibration strategy to the demander sub-model and the provider sub-model during the training of the joint model, avoids the intervention of the data provider data in the prediction stage through feature calibration, and solves the data security risk problem, effectively avoids unnecessary data flow, and reduces the data cost of the demander to a certain extent.

[0063] An embodiment of the present invention provides a non-volatile computer storage medium, which stores at least one executable instruction. The computer executable instruction can execute a federated learning method based on feature calibration in any of the above method embodiments.

[0064] The executable instructions can be used to cause the processor to perform the following operations:

[0065] Obtain a demander sub-model and a provider sub-model; the demander sub-model is trained by the business demander based on the business demander's data, and the provider sub-model is trained by the data provider based on the data provider's data;

[0066] Obtaining first feature data provided by a business demander and second feature data provided by a data provider, and training and generating a calibration model based on the first feature data and the second feature data;

[0067] The joint model is obtained by jointly training the feature calibration relationship of the calibration model, the demand sub-model and the supply sub-model;

[0068] In response to the data query request from the business demander, the first input feature data of the business demander is selected, and the second input feature of the data provider is determined based on the feature calibration relationship of the calibration model. The joint model is called based on the first input feature data and the second input feature to obtain the query result.

[0069] Figure 9The schematic diagram of the structure of the computing device embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0070] like Figure 9 As shown, the computing device may include:

[0071] Processor, Communications Interface, Memory, and Communication Bus.

[0072] The processor, communication interface, and memory communicate with each other via a communication bus. The communication interface is used to communicate with other devices, such as clients or other server network elements. The processor is used to execute programs, specifically, to perform the steps described in the aforementioned embodiment of a federated learning method based on feature calibration.

[0073] Specifically, the program may include program codes including computer operation instructions.

[0074] The processor may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the server may be processors of the same type, such as one or more CPUs, or processors of different types, such as one or more CPUs and one or more ASICs.

[0075] Memory is used to store programs. The memory may include high-speed RAM memory, and may also include non-volatile memory (non-volatile memory), such as at least one disk storage.

[0076] The program can be specifically used to cause the processor to perform the following operations:

[0077] Obtain a demander sub-model and a provider sub-model; the demander sub-model is trained by the business demander based on the business demander's data, and the provider sub-model is trained by the data provider based on the data provider's data;

[0078] Obtaining first feature data provided by a business demander and second feature data provided by a data provider, and training and generating a calibration model based on the first feature data and the second feature data;

[0079] The joint model is obtained by jointly training the feature calibration relationship of the calibration model, the demand sub-model and the supply sub-model;

[0080] In response to the data query request from the business demander, the first input feature data of the business demander is selected, and the second input feature of the data provider is determined based on the feature calibration relationship of the calibration model. The joint model is called based on the first input feature data and the second input feature to obtain the query result.

[0081] The algorithm or demonstration provided herein are not inherently relevant to any particular computer, virtual system or other equipment. Various general-purpose systems may also be used together with the teachings based on this. According to the above description, it is apparent that the structure required for constructing this type of system. In addition, the embodiment of the present invention is not directed to any specific programming language yet. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the above description of specific languages ​​is for the purpose of disclosing the best mode of the present invention.

[0082] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0083] Similarly, it should be understood that in order to streamline the present invention and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the invention, various features of the embodiments of the invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Accordingly, the claims that follow the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of the invention.

[0084] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition may be divided into multiple submodules or subunits or subcomponents. All features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed herein may be combined in any combination, except that at least some of such features and / or processes or units are mutually exclusive. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0085] Furthermore, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not other features, combinations of features from different embodiments are intended to be within the scope of the present invention and to form different embodiments. For example, in the claims below, any of the claimed embodiments may be used in any combination.

[0086] The various component embodiments of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It will be appreciated by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components according to an embodiment of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and a computer program product) for executing a part or all of the methods described herein. Such a program implementing the present invention can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0087] It should be noted that the above embodiments illustrate rather than limit the invention, and that alternative embodiments may be devised by a person skilled in the art without departing from the scope of the appended claims. In the claims, any reference signs placed between brackets should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising several different elements and by means of appropriately programmed computers. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments should not be understood as limiting the order of execution unless otherwise specified.

Claims

1. A federated learning method based on feature calibration, characterized in that: include: Obtain a demander sub-model and a provider sub-model; wherein the demander sub-model is trained by the business demander based on the business demander's data, and the provider sub-model is trained by the data provider based on the data provider's data; Acquire first feature data provided by the business demander and second feature data provided by the data provider, and train and generate a calibration model based on the first feature data and the second feature data; Performing joint training on the characteristic calibration relationship of the calibration model, the demander sub-model, and the provider sub-model to obtain a joint model; In response to the data query request of the business demander, the first input feature data of the business demander is selected, and the second input feature of the data provider is determined based on the feature calibration relationship of the calibration model, and the joint model is called based on the first input feature data and the second input feature to obtain the query result.

2. The method according to claim 1, characterized in that The training and generating of the calibration model based on the first feature data and the second feature data further includes: Using the first feature data and the second feature data as sample data, and dividing the sample data into training set data and validation set data; Performing training based on the training set data to obtain a calibration model; The calibration model is validated based on the validation set data.

3. The method according to claim 2, characterized in that The calibration model includes: a first calibration model; The training based on the training set data to obtain a calibration model further includes: Performing a clustering operation on the sample data in the training set data to obtain a first cluster of the business demander and a second cluster of the data provider; Extract the common feature data from the first feature data and the second feature data in the sample data, construct a cluster index based on the common feature data, and train and generate a first calibration model based on the cluster index. The cluster index is used to query the second cluster of the data provider based on the first cluster of the business demander.

4. The method according to claim 2, characterized in that The calibration model includes: a second calibration model; The training based on the training set data to obtain a calibration model further includes: The second characteristic data in the training set data is used as a prediction target, and a machine learning algorithm is used to perform prediction target fitting training based on the first characteristic data in the training set data to obtain a second calibration model.

5. The method according to claim 2, characterized in that The calibration model includes: a third calibration model; The training based on the training set data to obtain a calibration model further includes: Extracting relevant feature data between the first feature data and the second feature data in the training set data; The second feature data is used as a missing value, and data completion is performed on the second feature data according to the relevant feature data and a preset algorithm. The first feature data is used to perform feature calibration to train and generate a third calibration model.

6. The method according to claim 5, characterized in that The preset algorithms include: mean interpolation algorithm, maximum likelihood estimation algorithm and multiple interpolation algorithm.

7. The method according to any one of claims 1 to 6, characterized in that The calibration model includes: a first calibration model, a second calibration model and a third calibration model; The training and generating of the calibration model based on the first feature data and the second feature data further includes: Calculating prediction accuracy of the first calibration model, the second calibration model, and the third calibration model; The first calibration model, the second calibration model, and the third calibration model are sorted from high to low according to the prediction accuracy, and a final calibration model is determined according to the sorting result.

8. A federated learning device based on feature calibration, characterized in that: include: A sub-model acquisition module is used to acquire a demander sub-model and a provider sub-model; wherein the demander sub-model is obtained by training the business demander based on the business demander's data, and the provider sub-model is obtained by training the data provider based on the data provider's data; A calibration model training module is used to obtain the first feature data provided by the business demander and the second feature data provided by the data provider, and to train and generate a calibration model based on the first feature data and the second feature data; A joint model training module, configured to perform joint training based on the characteristic calibration relationship of the calibration model, the demander sub-model, and the provider sub-model to obtain a joint model; An application module is used to respond to the data query request of the business demander, select the first input feature data of the business demander, and determine the second input feature of the data provider based on the feature calibration relationship of the calibration model, and call the joint model based on the first input feature data and the second input feature to obtain the query result.

9. A computing device, characterized in that include: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to a federated learning method based on feature calibration as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that The storage medium stores at least one executable instruction, which enables the processor to perform operations corresponding to the feature calibration-based federated learning method according to any one of claims 1 to 7.

Images